Velocity-based training assessment: Effect of Extended-Relative-Phase-based extracted features on classification

Yeong Ju Woo, Hang Thi Phuong Nguyen, Hieyong Jeong

Keywords: relative phase, velocity-based training, human pose, feature extraction, Deep-learning-based personal training system

TL;DR: We proposed the Extended Relative Phase (ERP) indicator to evaluate velocity-based training using Human Pose Estimation (HPE).

Abstract:

Velocity-based training is one of the essential training methods that helps improve athletic performance by providing immediate feedback to athletes. However, there needs to be more ways to evaluate velocity-based training by analyzing the athlete's entire movement. Thus, this study aimed to verify the effectiveness of a newly proposed Extended Relative Phase (ERP) feature on the velocity-based training assessment by using the coordinates of most major joints using Human Pose Estimation (HPE). The difference between experts and novices was compared in the experiment using the proposed feature. The Relative Phase Angle ($RP_{Angle}$) exists to evaluate the combination of each joint's angular displacement and velocity. However, assessing the consistency of repeated movements and comparing angular displacement and velocity with experts takes work. For this reason, the Relative Phase Distance ($RP_{Distance}$) was proposed as a new feature.
The dataset trained and predicted the performance verification, including each joint angle, $RP_{Angle}$, and $RP_{Distance}$. The 1D CNN-based deep learning model for training and prediction was used to compare each extracted feature. As a result, the newly proposed indicator had a good effect on the prediction performance of the velocity-based training evaluation.

Submission Number: 16